1. 2013 II International Congress of Engineering
Mechatronics and
Automation (CIIMA)
23 al 25 de octubre del 2013
Universidad de La Salle- Bogotá Colombia
3. CONFERENCIAS
1. Diagnosis from Chronicles: an overview of related challenges
Audine Subias
CNRS; LAAS; 7 avenue du colonel Roche F-31400 Toulouse, France
Univ of Toulouse, INSA, LAAS, F-31400 Toulouse, France
subias@laas.fr
2. Integration of Different Facets of Diagnosis from Control and AI
L. Travé-Massuyés
CNRS, LAAS, 7, avenue du Colonel Roche, F-31400 Toulouse, France
Univ of Toulouse, LAAS, F-31400 Toulouse, France
Email: louise@laas.fr
PONENCIAS
1. Deducción y Validación de un Modelo Dinámico de la Transferencia Térmica en
un Invernadero a Escala
Oscar Alexánder Bellón Hernández
Facultad de Ciencias e Ingeniería
Universidad de Boyacá
Tunja, Colombia
2. Control Estadístico Aplicado a la Detección de Síntomas de Sucesos
Operacionales en Producción de Crudo con Sistemas de Levantamiento
Artificial BES .
Cesar Pereira, Jorge Prada.
Ecopetrol S.A.
Piedecuesta, Colombia
3. A comparative study of geometric path planning methods for a mobile robot:
Potential field and Voronoi diagrams
Edwar Jacinto Gómez, Fernando Martínez Santa, Fredy Hernán Martínez
Sarmiento. Distrital University Francisco José de Caldas
4. A comparative analysis of adaptive visual servo control for Robots Manipulators
in 2D
Maximiliano Bueno López, Daniel Mariño Lizarazo.
Ingeniería Eléctrica- Ingeniería en automatización- Universidad de La Salle.
5. Integración de redes de sensores inalámbricos (WSN) IEEE 802.15.4 – 802.11
para automatización industrial
Álvaro Romero, Alejandro Marín, Julián Orozco, y Jovani Jiménez.
Universidad Nacional de Colombia
4. 6. Filtro digital ajustable usando micro softcore en FPGA para frecuencias entre
200 hz y 20 khz
Jorge E. Reita, Juan C. Uribe, Edwar Jacinto, Fernando Martínez
Universidad Distrital Francisco José de Caldas,
7. Sistema Telemétrico De Registro De Señales De Emg Superficiales Basado En
Tecnología Bluetooth
Robín Alfonzo Blanco, Brian Chacón Hernández, Leonardo Andrés Góngora
Velandia.
8. Diseño De Un Equipo De Soldadura Basado En Gas HHO Extraído Del Agua
Gustavo Adolfo Ramírez Piedrahita. Facultad de minas.
Universidad Nacional de Colombia
9. Control GPI Multivariable de un Exoesqueleto para Asistencia de Marcha en
Personas con Discapacidad Motora
J. Arcos, A. Tovar, J. Cortés, H. Díaz, L. Sarmiento. Universidad de San
Buenaventura, Bogotá; Indiana University Purdue University Indianapolis;
Universidad Nacional de Colombia, Bogotá.
10. Detección y Seguimiento Facial en Niños Autistas con Bajo Nivel de
Funcionamiento
Y. Castro, J.C. Bejarano, J. D. Posada, J.A.Villanueva.
Department of Engineering, Universidad Autónoma Del Caribe –Barranquilla,
Colombia.
11. Modelo Para La Apliacion De La Norma Iec61131-3 En Un Sistema De
Manufactura Flexible
Oscar Mauricio Arévalo Rodríguez, Álvaro Antonio Patiño Forero
Universidad de La Salle
12. Diseño Scada de una autoclave, aplicando la norma IEC 61131-3.
Germán Alejandro Piñeros Bernal, Álvaro Patiño
Universidad de La Salle
13. Diseño y manufactura de un robot pendular suspendido de 4 gdl, para utilizarlo
dentro de un aplicativo de realidad virtual aumentada con colecciones de
museografia y telecontrol por internet
Méndez M. Luis Miguel, Uribe M. Bernardo y Pantoja R. Cesar Augusto.
Department of Mechanical and MechatronicsEngineering
Universidad Nacional de Colombia
14. Evaluación del Desempeño de un Controlador MPC para Una Planta
Multivariable de Tanques Interactuantes
Mario A. Zuñiga M., Pedro L. Rivera W., Francisco Franco
Departamento de Electrónica y Telecomunicaciones, Ingeniería en Automática
Industrial, Universidad del Cauca
5. Popayán, Colombia
15. Diseño E Implementación De Una Máquina Para La Producción De Papas
Chips
Leonardo Alberto Ciendua, Jonathan Jair Díaz, Luis Fernando Morales, Chiara
Seidel Schlenker, Álvaro A. Patiño-Forero, Jairo O. Montoya
Programa de Ingeniería en Automatización, Universidad de la Salle
Bogotá, Colombia
16. Tecnificación de equipos de control y monitorización de material particulado
para mejorar la calidad del aire en zonas de explotación y coquización de
carbón en Boyacá
Oscar AlexánderBellón Hernández, Dora Marcela Benítez Ramírez
Facultad de Ciencias e Ingeniería Universidad de Boyacá
17. Modeling and tracking control of a pneumatic servo positioning system
Iván Ramírez
18. Afinador de Guitarra Acústica Semiautomático
Juan Bejarano, Leandro Torres, Cesar Zúñiga, Facultad de Ingeniería,
Universidad Autónoma del Caribe
19. Analysis of alarm management in startups and shutdowns for oil refining
processes
Vásquez John, Prada Jorge, Agudelo Carlos, Jiménez José.Escuela
Colombiana de Carreras Industriales ECCI
Engineering department, Andes University
Automation Group Instituto Colombiano del Petróleo ICP
20. Optimización Multi-objetivo de un Controlador PID Aplicando Algoritmos Bioinspirados
Juan Camilo Castro Pinto, María Alejandra Guzmán Pardo
21. B-WalkMóvil -Sistema de Información Móvil para la Ubicación de Personas
Invidentes
Diana Lancheros Cuesta, Laura Cardozo, Laura Corredor, Ingeniería en
Automatización, Universidad de La Salle
22. Diseño y Construcción de un Prototipo de Máquina de Rehabilitación de Mano
y Muñeca
Camilo Andrés Cáceres Flórez, Jefry Anderson Mora Montañez, Robinson
Jiménez Moreno
GAV- Universidad Militar Nueva Granada
23. Comparación de enfoques de sistemas de control tradicionales y el paradigma
de los Sistemas Holónicos de Manufactura
6. Luis A Cruz Salazar, Oscar A Rojas Alvarado.
Universidad Antonio Nariño, Universidad del Cauca
24. Diseño Y Construcción De Un Rov Sumergible
Nelson O. Rodríguez Quiroz, Jairo O. Montoya G.
Universidad De La Salle - Programa de Ingeniería en Automatización
25. Implementación de redes neuronales y lógica difusa para la clasificación de
patrones obtenidos por un Sónar
Muñoz Aldana, David, Cruz Salazar, Luis A., Contreras Montes, Juan
Fundación Universitaria Tecnológico Comfenalco
Escuela Naval Almirante Padilla “ENAP”
26. Control de Temperatura en un Bioproceso Utilizando Lógica Difusa
Andrea Santos Morales, Cristian Camilo Beltrán Hernández, Claudia L.
Garzón-Castro
Grupo de Investigación CAPSAB- Facultad de Ingeniería - Universidad de La
Sabana - Chía, Colombia
27. Comparative analysis of non-linear filters for attitude estimation in a low-cost
inertial station
Sebastián López R., Julian,Munoz. Ruiz Fredy,Cheguini Mazeyar
28. Diseño de un Prototipo de Planta Para el Control de Nivel
Marco Tulio Calderón Acuña, Luis Hildebrando Alzate, José Ariel Gil García
Departamento de Física, Universidad de Caldas, Manizales, Colombia
29. Aproximacion Al Diseño De Los Eslabones De Un Robot Delta
Lucas Urrea Mantilla, Sergio Alejandro Medina, Ricardo Andrés Castillo, Oscar
Fernando Avilés
Programa de ingeniería en Mecatrónica, Universidad Militar Nueva Granada,
Bogotá, Colombia.
30. Diseño e implementación de un Prototipo de Torno Fresador de Control
Numérico Computarizado
Fausto Acuña
Departamento de Energía y Mecánica- ESPE, Latacunga, Ecuador
Andrés Gordón, Walter Núñez - Carrera de Ingeniería Mecatrónica- ESPE
Latacunga, Ecuador
31. Análisis Comparativo De Técnicas De Control Convencional E Inteligente Con
Los Sistemas De Articulación Flexible Y Bola Biga
Germán E. Polanco Aristizábal, Oscar E. Soto Castañeda, Jesús A. López
Departamento de Automática y electrónica, Universidad Autónoma de
occidente.
7. 33. Diseño de Control Neuronal por PLC para una Planta de Laboratorio
Mario A. Fernández F., Universidad de Talca, Curicó, Chile
William Gutiérrez M., SENA-Regional Valle, Cali, Colombia
Jesús A. López S., Universidad Autónoma de Occidente, Cali, Colombia
32. Inteligencia De Enjambres Aplicada Al Control Adaptativo
Navas, Andrés Felipe., López, Jesús A.
Universidad Autónoma de Occidente – Cali
33. Implementación de una Red Neuronal Artificial tipo SOM en una FPGA para la
resolución de trayectorias tipo laberinto
Callejas Iván, Piñeros Juan, Rocha Juan, Hernández Ferney, Delgado Fabio
Ingeniería Electrónica, Universidad INCCA de Colombia
Bogotá, Colombia
34. Reconocimiento de Embarcaciones Marinas Usando Redes Neuronales
Esmeide Leal, Nallig Leal, Ronald Messino, Richard Aroca.
35. Cálculo de camino óptimo para manipulador articulado SCARA sujeto a
obstáculos.
Carlos G. Pillajo
Departamento de Control y Automatización
Universidad Politécnica Salesiana
Quito, Ecuador
36. Estimación de orientación de herramienta y tuerca utilizando la visión del robot
NAO
Carlos Peña, José Hoyos, Flavio Prieto Jorge Ayala, Claudia Garzón-Castro
Facultad de ingeniería
Universidad Nacional de Colombia, Universidad de La Sabana
37. Diseño de un Sistema de Maniobra y Pateo la Bola, para un Robot Categoría
SSL para la RobocuP
Rangel Díaz, Jorge Eliécer, Sanabria Torres, Jairo Andrés
Universidad de La Salle; Trane de Colombia SA
38. Técnicas De Control Adaptativo Aplicadas A Un Mezclador Por Baches Con
Agitación Continua
Juan Esteban Betancur, David Velásquez Rendón, José Fernando Osorio
Brand, Rigoberto Maldonado.
Escuela de Ingeniería de Antioquia. Envigado
39. Design and implementation of a prototype orthoses for prevention and
treatment of Carpal Tunnel Syndrome (CTS)
8. Karen Edilma Garzón Cruz, José Luis Rubiano Fernández
Universidad de la Salle, Bogotá
40. Design of an under actuated Altering tetrapod Robot bio Inspired on Scorpions
Jorge I. Montalvo N. Universidad Autónoma de Occidente
POSTER
1. Sistema automatizado para la caracterización de la calidad espacial de un haz
láser
Norma Alicia Barboza Tello, Eduardo Antonio Murillo Bracamontes, José Luis
Rodríguez Verduzco
2. Control Basado en PLC de un Brazo Robótico para el Transporte y
Almacenamiento de Productos en una Celda de manufactura
Jurado Muñoz Sandra M., Cubillos Rojas Jean A., Muñoz Magín Elviz Jhony,
Muñoz Tafur Johan Fabián. Institución Universitaria Tecnológica de
Comfacauca Unicomfacauca
Facultad de Ingeniería – Tecnología en Electrónica- Popayán, Colombia
3. Desarrollo De Una Herramienta Extrusora De Polímero Utilizada En Una
Impresora 3d Fdm
Edgar A. Torres, Jersson X. León, Edwin Torres. Escuela de Ingeniería
Electromecánica, Grupo de Energías y Nuevas Tecnologías-GENTE
Universidad Pedagógica y Tecnológica de Colombia, Duitama, Colombia
4. Sistema Domótico Para Discapacitados Controlado Por Voz
Iván Santiago García Peñaloza, Luis Eduardo Sierra Catillo, Adriana Patricia
Arias Díaz, Edwin Ferney Bonilla Torres, Salvador Pacheco. División de
Ingenierías y Arquitectura, Facultad de Ingeniería Mecatrónica, Universidad
Santo Tomás- Bucaramanga, Colombia.
5. Control Adaptativo en una junta P-R (del tipo Prismático Rotacional)
Gamba, Nicolás. Sierra, Nelson y Romero, David. Universidad Nacional de
Colombia
6. Modelo Para La Implementación De La Norma Iec 61131-3 En Un Sistema
Integrado De Manufactura
Julio Alberto Ambrosio, Erick Stiven Ariza, Julián David Guaqueta, Álvaro
Antonio Patiño. Universidad de la Salle, AVARC/SAVARC
Bogotá D.C, Colombia.
7. Plataforma didáctica para el estudio de procesos térmicos en laboratorio de
Instrumentación industrial
9. A. Chacón García, H. Montaña Quintero, Departamento de Ingeniería
Electrónica Pontificia Universidad Javeriana, Bogotá, Colombia
Tecnología en Electrónica Universidad Distrital Francisco José de Caldas
8. Actuador Hidráulico para Prótesis de Rodilla
Edilberto Mejía Ruda, Sebastián Jiménez Gómez, Oscar Fernando Avilés
Sánchez, Oscar Iván Caldas Flautero, Juan Camilo Hernández. Mejía,
Programa de Ingeniería Mecatrónica –Universidad Militar Nueva Granada
9. Analysis of Kinematics and Dynamics for ABB IRB-140 Serial Robots and
Evaluation of Energy Consumption in the Tracking of a Path.
Mauro Baquero Suarez, Ricardo Ramírez Heredia, Mechatronics Engineering
Department, Universidad Nacional
10. Control De Plataforma De Stewart Mediante Procesamiento De Imagen
Robinson Jiménez Moreno, Oscar Fernando Avilés S. y Jorge Riveros,
GAV - Universidad Militar Nueva Granada
11. Diseño de un Sistema Dosificador y Mezclador de Concentrados
H. González, H. González, J. Bohórquez, J. Quintero, J. Gómez
Grupo de Investigación de Control & Mecatrónica - UNAB
12. Diseño De Controladores Pid Para Sistemas De Segundo Orden Usando
LoopShaping Robusto
Mario F. Jiménez, Andrés A. Ramírez.
Departamento de Ingeniería Mecatrónica - Fundación Universitaria Agraria de
Colombia.
13. Desarrollo de un Sistema SCADA inalámbrico con Zigbee y Arduino
Herrera Jean, Barrios Mauricio y Pérez Saúl.
Programa de Ingeniería Mecatrónica, Universidad Autónoma del Caribe
14. Control remoto de un robot móvil LEGO Mindstorm mediante Visión por
Computador
German Andrés Rivas Lema, Andrés Felipe Gálvez Leyes, Jimmy Alexander
Cortes Osorio
Universidad Tecnológica de Pereira.
15. Investigación De Nuevos Modelos Nanotecnológicos En El Diseño De Piel
Artificial Con Nanoinstrumentación Fabricada Por Electrospinning Para El
Recubrimiento De Prótesis De Mano Y Pierna En Discapacitados
Antonio Faustino Muñoz, Aldo Pardo García. Universidad Autónoma de
Bucaramanga – Grupo de Investigación de Control & Mecatrónica, Universidad
de Pamplona Instituto de Investigaciones Tecnologías Avanzadas IIDTA,
Universidad del Cauca, Grupo en Automática Industrial
10. 16. Dispositivo Traductor Del Lenguaje De Señas De Personas Sordas A Sonidos
Auditivos De Las Letras Del Abecedario.
Carrasco Harold, Encalada Lennin, Universidad Técnica del Norte, Ibarra Ecuador
17. Sistema de seguridad para la conducción de vehículos mediante el análisis
facial de una persona utilizando visión artificial
Alvaro Fuentes, Gabriela Estrella, Carlos Acosta, Juan Nazate,. Carrera de
Ingeniería Mecatrónica, Universidad Técnica del Norte, Ibarra, Ecuador
18. Reconocedor Facial Usando PCA y Redes Neuronales
Manuel Alejandro Díaz granados Santos, Universidad Autónoma de Occidente,
Cali, Colombia
Jesús A. López, Universidad Autónoma de Occidente, Cali, Colombia
19. Diseño e Implementación de un Sistema Bio-inspirado Para la Simulación del
Depredador y la Presa Implementado sobre Plataformas Lego
Kristel Solange Novoa Roldán, Héctor Iván Tangarife Escobar, Rhonier Ernesto
Machado Mosquera
Universidad Distrital Francisco José de Caldas, Facultad Tecnológica.
Grupo de Investigación Robótica Móvil Autónoma- ROMA
20. Plataforma De Entrenamiento En Tareas De Telecirugía
Samuel Quintero M , Oscar Avilés S, Darío Amaya, Robinson Jiménez.
Grupo de investigación GAV - Universidad Militar Nueva Granada
21. Diseño y Construcción de una Máquina Llenadora Semiautomática para Bolsa
Plástica
González Barreto Sergio Fabián, Rodríguez Ardila Julián Felipe
Tecnología Mecatrónica, Escuela Tecnológica Instituto Técnico Central
22. Diseño de dispositivo para el control de Multímetros Fluke Serie 80
William Darío Aguirre Hernández, Diego Alejandro Fajardo Vargas, Fabio
Lorenzo Roa Cárdenas
Ingeniería Mecatrónica, Fundación Universitaria Agraria de Colombia
23. Segway Plataforma De Un Grado De Libertad
Kristel Solange Novoa Roldán, Mauricio Diusaba Rodriguez, Yesid Urueña
Cuervo
Universidad Distrital Francisco José de Caldas, Facultad Tecnológica.
Grupo de Investigación Robótica Móvil Autónoma- ROMA
11. Diagnosis from Chronicles: an overview of related
challenges
Audine Subias
CNRS; LAAS; 7 avenue du colonel Roche F-31400 Toulouse, France
Univ of Toulouse, INSA, LAAS, F-31400 Toulouse, France
subias@laas.fr
-
Abstract—Chronicle recognition is an efficient method to
address the problem of diagnosis and more generally the problem
of situation recognition. Several researches have investigated this
direction to develop approaches for dynamic complex systems.
But chronicle recognition gathers other interesting research topics
related notably to the field of machine learning and to timed
transition systems modeling. This article gives a picture of
different theoretical and applicative works connected to chronicle
recognition which is an active research area.
Index Terms—Diagnosis - Chonicle recognition- Diagnosability
analysis - Chronicle learning
-
event(E,t): an event type E is stamped with t the date
of its occurrence.
noevent(E, [α, β]): this predicate defines a forbidden
event. No event E occurs between α and β time units.
occurs((m, n), E, [α, β]) : at least m and at most
n occurrences of an event E between α and β time
units.
A notion of domain attribute is also defined by a couple
E : e where E is the attribute name and e is a possible value
of the attribute. The set of possible values defines the domain
of the attribute. A domain attribute as a unique value at each
time instant t. In this way, the predicate event(E : (e1, e2), t)
models a change in the value of domain attribute E from e1
to e2 at time t and the predicate noevent(E : (e1, e2), [α, β])
forbids the change of value of E between α and β time units.
Finally, a set of actions can be launched and some events
can be emitted when a chronicle is recognized. Fig 2 gives two
simple examples of chronicles according Dousson’s language
description.
I. C HRONICLES WORLD
A. What are chronicles ?
Most of the works on chronicles are issued from the French
community. [29] has initially developed this model to capture
automatically the evolutions or partial evolutions of dynamic
systems. The evolutions to monitor are described in terms of
temporal patterns called chronicles. A chronicle is not a simple
execution trace of the system it is a discriminant observable
part allowing to recognize a particular situation. Chronicles are
expressed in a specific language and then translated into time
constraints satisfaction graphs. The nodes of the graphs are
associated to the events, and the edges are labeled by the time
constraints (see Fig 1).
Chronicle SequenceAB {
event(A, t1)
event(B, t2)
0 < t2 - t1 < 2 -- sequence within 2s
Chronicle Noevent_In_AB {
event(A, t1)
event(B, t2)
noevent(C, (t1 t2)
0 < t2 - t1 < 2 -- sequence within 2s
when recognized emit event(C, t2)
}
[4, 6]
[1, 3]
B
A
[2, 5]
D
when recognized emit event(D, t2)
}
Fig. 2: Chronicle of sequence AB in [0, 2] (left) and no event
C in AB (right).
]0, +∞[
C
Chronicle based approaches can be related to other methods
to represent situations stressing on the temporal dimension
such that situation calculus introduced by [55], the event
calculus [51] or the temporal interval of Allen [5],[6]. All these
methods are commonly used in the Artificial Intelligence field
for representing and reasoning about temporal information. One
major advantage of chronicles compared to these approaches
is the rich formalism allowing one to describe the observable
patterns corresponding to behaviors one wants to detect. In
particular, chronicles account for partial orders between events
easily and are also able to the lack of events via forbidden events.
Another advantage lies on the efficiency of the recognition
Fig. 1: A chronicle
This kind of approach assumes that a time stamp or
occurrence date can be assigned to each event. A chronicle is
therefore a temporal pattern described in terms of events and
time constraint between event occurrence dates.
In [29] the chronicle language is based on the notion of
predicate. A predicate defines the events required for the
recognition and the events which must be discarded. A chronicle
is recognized if all the predicates are satisfied. The major
predicates that have been defined are:
1
12. system which makes chronicles suitable for real-time operation
(see section I-B). The main drawback of the chronicle based
approach is the design of the chronicles. How acquiring and
updating the chronicles? We will see in section III that several
approaches have been proposed to remedy this design problem.
Chronicle recognition consists in identifying in an observable
flow of events all the instances of the chronicles i.e. all possible
matchings between an input flow of events and a chronicle. The
identification is performed on the fly, as soon as the events occur.
When a new event occurs it is integrated into the chronicle if it is
consistent with the expected event of the pattern and if its time
stamp is consistent with the time constraints of the chronicle.
Each new instance of chronicle generated is a new hypothesis
and added to the set of hypotheses. The chronicle recognition
system must then manage on-line all these instances i.e. all the
hypotheses elaborated in time. Instances are discarded when
time constraints are violated. Finally, a chronicle is recognized
when a complete match is observed. For one given flow of
events multiple instances of a chronicle can be recognized
in a sequential way or simultaneously. For example, let us
consider the simple chronicle defined using the description
language previously presented: event(A, t1 ) ∧ event(B, t2 ) ∧
event(C, t3 ) ∧ (t2 − t1 ) ∈ [1, 3] ∧ (t3 − t2 ) ∈ [0, +∞[: an
event A occurs followed by an event B with an interval
of 1 to 3 time units. B is then followed by an event C
(see Figure 3). Let us consider the observed event flow:
(A, 0), (B, 2), (A, 8), (B, 9), (C, 11). At t = 11, two instances
of this chronicle are simultaneously recognized when the event
C occurs (see Figure 4). Note that, in this example the two
partial instances (A, 0), (B, 2) and (A, 8), (B, 9) will never be
discarded as the event C can occur whenever and the time constraint will never be violated. With an other observed event flow
given by (A, 0), (B, 2), (C, 8), (A, 9), (B, 10), (C, 30) two instances of the chronicle would be recognized in a sequential
way.
A
[1, 3]
B
C
Fig. 3: A simple chronicle
[29] has developed a Chronicle Recognition System (CRS)
that performs an exhaustive recognition.
Another chronicle based approach has been developed by
[16] providing also a chronicle recognition system called
CRS/ON ERA designed on the basis of duplicating automata
and able to detect on line chronicle instances (see section IV-D).
The main difference between the two approaches is the syntax
used to describe the chronicles. Another difference concerns
the way the events are managed by the chronicle recognition
system. In CRS/ON ERA the events are managed according
a first in first out mechanism whereas in CRS the events
are managed according to their occurrence dates. The two
systems are implemented with object languages and can be
integrated as libraries with other applications or used via an
Instances
0
{(A,0)}
2
{(A,0)},{(A,0),(B,2)}
8
{(A,0),(B,2)},
{(A,8)}
9
{(A,0),(B,2)},
{(A,8),(B,9)}
11
B. Chronicle recognition
Time
{(A,0),(B,2)},
{(A,0),(B,2),(C,11)},
{(A,8),(B,9)},
{(A,8),(B,9),(C,11)}
13
{(A,0),(B,2)},
{(A,8),(B,9)}
Fig. 4: Instances of chronicles
independent executable. More recently a new recognition tool
called Chronicle Recognition Library (CRL) based on the
semantic of the chronicle language CRS/ON ERA has been
proposed [17].
Chronicle recognition can be related to other situation recognition systems like the temporal diagnosis system developed
by [36] for tracking hepatitis symptoms, the work of [57] for
planning and matching an observation sequence applied to
the diagnosis of CN C-machining centers and also the work
of [43] to diagnose trends in growth patterns of pediatric
patients. The approach proposed in [49] illustrated in the
domain of driverless transport systems must also be mentioned.
The difference between these systems concerns mainly the
temporal framework used and their ability to represent complex
temporal behaviors and complex structured situations. Another
relevant criterion is obviously the computational efficiency of
the situation recognition system. Plan recognition is another
research area that can be related to chronicle recognition. In
the context of multi agent systems plan recognition consists in
deriving the underlying plan executed by an agent based on
partial observation of its behavior. The main difference is that
plan recognition generally focusses on the composition of the
plan in terms of actions rather than on time aspects. Moreover,
the context (goals, preferences and capabilities of the agent,
effect of the plan execution...) in which the plan is generated
is also considered [60].
II. D IAGNOSIS BASED ON CHRONICLE RECOGNITION
When one use chronicle for diagnosis purposes there are
two main ways to consider problem. Chronicles can model
the normal behavior of the system one wants to diagnose. The
diagnosis problem is then tackled as a consistency problem
between the observations and the model of the system. In this
case, the chronicle recognition allows to detect any discrepancy
between the normal behavior of the system and the real behavior
given through the observations (that are supposed safe). Another
possibility is to consider chronicles of faulty behaviors. The
efficiency of such an approach relies on the direct link between
the symptom of a fault and the fault itself. Nevertheless, it
differs from classical abductive diagnosis systems as time
aspects are dominant. Generally diagnosis applications based
on chronicles need not only the two types of chronicles (normal
and faulty) but also a real chronicle base, what is not trivial to
design (see section III).
13. AUTONOMOUS COMMUNICATING SYSTEMS
A. Use of chronicles for diagnosis
The chronicle approach has been developed and used in a
wide spectrum of applications [22]: for telecommunication
systems to manage alarms [27] or in production domain
to monitor gas turbines [2]. In the medical field also, for
cardiac arrhythmia detection [20], where electrocardiogram
interpretation is made by chronicles: a symbolic description
with time constraints is associated to pathological situations.
In [52] chronicles are used to alarm processing in power
distribution systems. More recently chronicles have been used
in intrusion detection system. In [56] a chronicle approach for
alarm correlation is proposed. Chronicles are not directly used
to describe attacks but to represent known phenomena which
involve several alarms. Chronicles are used both to represent
normal phenomena allowing to discriminate legitimate actions
from attacks and malicious (and deterministic) phenomena
involving many events are involved. In video understanding and
more precisely in the context of visual monitoring applications
for human security purposes, a formalism very closed to
chronicles is proposed to describe the concepts involved in
activity recognition. The objective is to detect suspect human
behavior operators [62]. In the field of unmanned aircraft
systems U AS, chronicles have been introduced for handling
breakdowns and to check the consistency between the activities
in U AS [19] but also for the successful deployment of a fully
autonomous unnamed aerial vehicle operating over road and
traffic networks by detecting vehicles overtaking and passing
other vehicles [44]. In the context of high level architecture
simulations [11], chronicle recognition is integrated into the
development of a simulation as a component to analyze on line
the data. Another important field of application of chronicle
recognition is collaborative systems notably web services
[23]. In this case the main challenge is the distribution of
the chronicles into subchronicles and the communication or
synchronization mechanisms between the chronicles [13],[40].
The next section gives an overview of two recent projects
benefiting from chronicle recognition.
B. Chronicles and collaborative communicating systems
More and more systems take benefit of communication
supports and achieve their objectives in a networked distributed
and cooperative way. For these systems, coping with context
changes requires considering self-adaptive communication
protocols in the design step so that the communication system
configuration then dynamically changes according to the user’s
requirements and to the load of the communication resources.
Dealing with this problem requires the capacity of detecting
the possible degradations of the Quality of Service (QoS) and
of dynamically modifying the behavior of the communication
protocols for each new context situation. This requires in turn
both monitoring the QoS values, detecting the degradations,
identifying their origins through appropriate diagnosis and executing reconfiguration actions. The DAISY project (Diagnosis
for AdaptIve Strategies in collaborative sYstems) tackles the
problem of providing adaptability to the traffic control and
management system [65]. DAISY focusses on adapting the
which mo
and relat
observing,
on the oth
Coping with context changes in networked systems requires to provide adaptability to the traffic control and
management system. This can be achieved through selfcommunication protocols at the transport level for coping readaptive communication protocols that dynamically with 3. SITUA
the dynamically changing contextsystem according from the
configure the communication situations arising to the
distributed and collaborative mobile applications.
user’s requirements and to the load of the communication
resources.
A taxonomy of transport services provided by the existing 3.1 Princ
transport protocols has beenSystem Interconnection showing
In the well-known Open elaborated [26] [30], (OSI)
that existing transport protocols present several limitations with In the pr
referential model (Zimmerman (1980)), composed of seven
layers, the application requirements. It layer operating on
regard to QoStransport layer is the lowest is important to notice alerting th
an end-to-end basis between two based on implementations
that most of the transport services areor more communicating a new situ
hosts. This layer is located between the applications error
where mechanisms offering different functionalities (i.e.and the new si
the network layer. Transport services enable applications the QoS p
control or congestion control) are merged within the same Therefore,
to abstract the communication services and protocols promonolithic implementation. Such a solution has a limited scope to detect
vided by the lower network and MAC (Media Access
of Control) layers. Transport protocols specifyhandling mech- discrimina
applicability and assumes a predefined QoS the mechaanisms already implementedintegrated during the design-time situations
nisms to be known and in order to offer the required
of transport services. Because theTherefore [67] [66] propose The entire
the communication protocol. communication Quality of
Service (QoS) is highly impacted by the with a composition analysis o
to perform the orchestration of the traffic specific transport
protocol in use, our self-adaptation architecture targets
of components that provide different and well-identified QoS communic
the transport level and proposes to adapt the transport
properties to the traffic [30]. This context evolves. approach time serie
protocol as the communication component-based
be pointed
resulting of the combination of pluggable components offering
Dealing with this problem not facilitate the a proper
specific functionalities, can widely only requires design and
characterization of the alternative protocol properties but
development of new composed transport services. assess the
also the capability of monitoring the QoS to
The DAISY context. These are at the basis of the
communicationproject suggests the use of chronicles associated to the different relevant modifying the behavior of theto
decision to dynamically traffic situations to be detected
communication protocol for each new context situation
guide the composition of these pluggable components. In other
and the objective appropriate reconfiguration actions.
words, executing theof DAISY is to guide the reconfiguration
strategies. 1 illustrates the architecture [3] foreseen to provide
Figure The proposed architecture [4], that has been proposed in (Aguilar-Martin given Figure .
a solution to this problem is et al. (2011)), 5and that is foreseen to provide a solution to this problem.
• an off
analy
ficati
syste
teriza
terns
corre
syste
lows
of hig
event
the t
sema
tions
• an on
tified
deter
acqui
the s
Reco
ure 1
This pape
following
(1) Gene
comm
(2) Speci
The Reconfiguration/Decision System outputs the protothis i
col to be deployed. This decision is taken upon several
The Reconfiguration/Decision System outputs the protocol
is ass
inputs:
to be deployed. This decision is taken upon several inputs:
A cla
• the properties the of the different available protocols
Po different available protocols gathered
comm
• the properties of
gathered through an ontology
fore t
through an ontology
• the communication context at time t0 Cx (t0 )
chron
• • the properties P context at time 0
the communication required by the application and the
a
seque
• the properties context Ca by the application
application Pa required
Each
• • the current context Cx (t) recognized by the Context
the current context recognized by the Context Recognition
of sa
ˆ
tive p
Recognition System, i.e. Cx (t).
System
Fig. 1. Architecture for self-adaptation
Fig. 5: Architecture for self-adaptation
The Context Recognition System monitors and assesses the
communication context and related QoS, receiving information
from monitors observing, on one hand, the application context
Ca and, on the other hand, the network context Cx . The mission
of the Context Recognition System is to perform situation
recognition; this means to supply to Reconfiguration/Decision
14. System relevant information every time a new situation arises
on the network and also to identify this new situation.
A situation is related to an evolution of the QoS parameters
of the studied communicating system. Therefore, situation
recognition induces the capability to detect different relevant
traffic situations taking the discriminating features in terms
of QoS indicators of such situations into account. Situation
recognition strategy relies then on the analysis of available
information issued from the communicating system and is
based on two different but non-independent steps:
• an off-line step in which historical data are analyzed and
processed to characterize the known behavior of the system
in terms of chronicles. This step constitutes a learning
stage. Learning methods such the one presented in III-B
are investigated during this step. The events involved in
the chronicles arise from feedback provided by standard
parameters stamping the packets.
• an on-line step, during which the system characterization
and the on line data are used to determine the current
expected state of the process. This stage is equivalent to
a chronicle recognition step.
C. Chronicles and service oriented applications
Service Oriented Architecture (SOA) is a software development model in which an application is broken down into small
units, logical or functional, called services. SOA allows the
deployment of distributed applications very flexible, with loose
coupling among software components such as web services,
which operate in heterogeneous distributed environments. The
services are inherently dynamic and then cannot be assumed
to be always stable as the resulting service can be altered
by external events such as changes in interfaces, misbehavior
during operation etc. In the case of service composition the
failure of a single service leads to error propagation in the
other services involved, and then to the failure of the whole
system. Such failure often cannot be detected and corrected
locally (into a single service). It is then necessary to develop
suitable architectures for the diagnosis and the correction of
failures, both at individual (service) and global (composition)
level. To face this problem, an adaptive architecture managing
SOA has been proposed in [69]: ARM ISCOM (Autonomic
Reflective MIddleware for management Service COMposition)
is a Reflective Middleware Architecture. The reflection is
the ability of ARM ISCOM to monitor and change its own
behavior, as well as aspects of its implementation (syntax,
semantic, etc.), allowing the ability to be sensitive to its
environment. Based on autonomic computing ARM ISCOM
provides an autonomic manager and an ontology framework to
manage all the knowledge used by the middleware as depicted
in Figure 6. The autonomic manager implements the intelligent
control loops according a M AP E (Monitor, Analyse, Plan,
Execute, Knowledge) schema [45]. The diagnoser module
of this architecture relies on a chronicle recognition system
to diagnose failures of the services. The system behaviors
associated to these failures are then represented by chronicles
designed from the events involved in the choreography of
Fig. 6: MAPE loop in the middleware architecture ARMISCOM
the services. The monitor component sends to the diagnoser
information related to the event occurrences, which allows the
diagnoser to recognize the chronicle instances and to detect
and identify failures. To provide a fully diagnosis architecture a
diagnoser is associated to each software component of the SOA
application. The distribution of chronicles into subchronicles
is then one of the problem addressed in this work. Into each
diagnoser, a subchronicle is a local chronicle and is defined as a
subset of events and/or with a less-constrained time constraints
graphs. Elements for synchronization have been introduced
into (sub)chronicle definition so that the chronicle recognition
process is totally distributed [70]. Each local chronicle involves
specific events called linked events that establish the connexions between the subchronicles. Additionally, each event has
attributes from which the consistency of the whole operating
is checked. ARM ISCOM is deployed in the context of web
services in a classical shop application. Two main types of
failure are investigated: the violation of an acceptance of service
from the warehouse and a delay in the deliverance of a service.
The first type of failure is local to a service whereas the second
one traduces a failure in the choreography.
III. C HRONICLES DESIGN
A. Chronicles design methods
Diagnosis from chronicles must deal with the problem of
the chronicle design as model based diagnosis approaches have
to face the model building. Model based chronicle generation
approaches have been developed. For instance, in [39] the
runs of the monitored system are described in the temporal
tiles formalism. The authors propose an algorithm inspired
of Petri net unfolding to build all the temporal runs of the
system. Then, the projection of these runs on the observable
part allows to define the chronicles. Other approaches have
been investigated from learning theory. One can consider
for instance learning techniques based on Inductive Logic
Programming (ILP) [54] [20], case-based chronicle learning
of [33],[34] that is a characteristic supervised method by
reinforcement learning but also [41],[71],[42] that adapt a
15. clustering method to learn chronicles in an unsupervised way by
projecting chronicle instances into a normative space. Finally,
chronicles are also acquired from approaches that analyze logs
and extract the significant patterns by temporal data mining
techniques. In [32],[31], the chronicle learning problem is
motivated by discovering the most frequent alarm patterns in
telecommunication alarm logs and their correlations. The tool,
called FACE (Frequency Analyzer for Chronicle Extraction),
extracts the frequent patterns by carrying out a frequencybased analysis on sublogs, defined on windows of time of
fixed duration. The frequency criterion is defined as a userdefined minimal frequency threshold. It is important to notice
that the considered patterns refer to a single event. For one
event, the alarm log is represented as an histogram of the
number of occurrences of the considered event in each time
window. Then, a normalization transforms this histogram into a
cumulative graph that depicts the sum of the occurrences from
the beginning, normalized by the sum of occurrences during
the sub-log corresponding to the current time window. Finally
a Self Organizing Map (SOM) algorithm [50] is applied to
the set of graphs to aggregate different temporal patterns and
detect correlations.
B. Focus on a chronicle learning approach
In the context of the DAISY project that we have presented
in Section II-B, the design of the chronicles that model
the communicating system situations is a real challenge. An
approach has been proposed to learn these situations at the
transport level relying on data mining techniques [35][4]. The
objective of temporal data mining techniques is simply to
discover all patterns of interest in the input data, which is an
unsupervised and explanatory task. There are several ways to
define the interestingness of a pattern. However a frequency
criterium is widely used [28][68][24] for unearthing temporal
patterns. Among existing approaches, one can distinguish two
main temporal pattern discovery frameworks: frequent episodes
[53] and sequential patterns [1]. Frequent episode framework
uses a single and long sequence and copes with the discovery
of temporal patterns called episodes that occur sufficiently in
the sequence. An episode is a partially ordered set of events.
The notions of frequent episode and subepisode are defined.
In [53] episode discovery focusses mainly on two types of
episodes: serial episode when the order between the events is
total and parallel episode when there is no order along the events.
Sequential patterns framework is based on the discovery in a
collection of sequences of all possible time ordered itemsets (i.e.
sequence of items) with sufficient support w.r.t the user-defined
threshold. The support of an itemset is defined as the number
of time the item can be observed in the collection. Further, a
sequence is said to be maximal in a set of sequences, if it is not
contained in any other sequence. Sequential pattern discovery
relies then on the systematic research of maximal sequences
that have a support at least equal to the minimal support userdefined. Many methods of search for sequential patterns are
designed along the same lines as the Apriori algorithm [1].
For the problem of traffic assessment on great dynamic
systems addressed in DAISY , a given situation can be
induced by several scenari. For instance, for a fixed network
topology and fixed transport level protocols several situations
of congestion can occur. If from a static point of view the
features of these congestions are similar from a dynamic point
of view they are different. It is then necessary to take several
scenari of a same situation into account for the learning of the
temporal patterns. Therefore, the chronicle discovery approach
proposed allows to discover frequent chronicles from multiple
sequences that is to say the chronicles that are frequent in each
sequence and not on only on the collection of sequence as
in sequential pattern approaches. Moreover, the proposal is to
discover the chronicles not only for a given frequency criterium
but for all the possible frequencies. The goal of the proposed
algorithm [35] is to build large frequent chronicles in an event
log composed by multiple runs. The chronicles are built by
incrementation. At each step, candidate chronicles of size i are
built from frequent chronicles of size i − 1 and are kept if they
are frequent as well [24]. A frequent chronicle is a chronicle
for which the frequency of appearance in the sequence is larger
than a user defined threshold. When a larger frequent chronicle
cannot be found, the search is stopped. In the final set of kept
chronicles, there is no chronicle that can be a subchronicle of
another one. This means that the algorithm returns the longest
frequent chronicles in the event flow.
IV. C HRONICLES AND OTHER FORMALISMS
Chronicles are temporal pattern and then can be viewed
as timed transition systems. From a diagnosis point of view,
the interest is as we will see in section V to apply well
known approaches of diagnosability analysis developed on
such systems. By using timed transition systems such as Time
Petri Nets (T P N ) [59], one can take benefit of the different
methods of model checking to explore the state space of the
system and to check established properties. Thus, several works
have been developed to translate or to transform chronicles.
A. Chronicle modeling based on Time Petri Nets
Time Petri Nets (TPNs) is a prominent tool to model timed
discrete event systems as several effective analysis methods have
been proposed [59][9]. T P N s extend Petri nets with temporal
intervals associated to transitions. Firing delay ranges are
associated to transitions. The work developed by [63] focusses
on the transformation of chronicle to T P N models. They
consider specific chronicles called Causal Temporal Signature
(CT S) that are expressed by a conjunction of triples (A, B, T )
where A and B are events and T a time constraint. The authors
consider three types of time constraints which are defined by
interval structure: date, time and duration. A date structure is
used to model the time between the occurrence of two events
whereas a delay structure allows to model the time between
the occurrences of two events taking into account a degree
of uncertainty. Finally, the duration structure is similar to a
”hold” predicate as it is used to express that information is
true from a date and remains true throughout the time interval
involved. The problem addressed in this work concerns the
17. marking m, firable instantly and if no transition with higher
priority satisfies these conditions. The authors formally define
in the T P N P r formalism several basic patterns from which a
chronicle model can be composed. These patterns correspond
to common structures in a chronicle such that:
-
event(a, t) ∧ t ∈ [α, β[:] an event a occurs between
α and β time units;
event(a, t) ∧ noevent(b, [0, t[):] an event a occurs
without any prior event b;
event(a, t) ∧ occurs((m, n), b, [0, t[):] an event a
occurs after at least m and at most n events b;
Fig. 9 gives the LT P N P r of a chronicle based on the two
predicates: event and occurs. The chronicle is recognized if
the “pok ” places is marked. Initially, the place pinit is marked
and m events of type b are expected to fire the transition t2 .
Then, if at least n − m + 1 events b occur — i.e. a total of at
least n + 1 events b — the chronicle is not recognized. On the
other side, if an event a occurs before the n − m + 1 events
b, then the chronicle is recognized. Finally, priorities (dashed
edges) ensure compliance of the LT P N P r with the chronicle
bounds.
pinit
t1
b
p1
[0, 0]
m
t2
b
p3
[0, 0]
p4
t4
n−m+1
tok
a
pok
Fig. 9: event(a, t) ∧ occurs((m, n), b, [0, t[)
Three types of systematic combination templates of these
patterns are also considered:
-
D. Chronicle Recognition modeling based on Coloured Petri
Nets
The objective of the work [10][11][18][21] is for a given
chronicle to establish in a progressive way the list of all
its recognitions. A chronicle is defined as a single event,
the conjunction of two chronicles, the disjunction of two
chronicles, the sequence of two chronicles or the absence
of a chronicle during another chronicle. This is noted by:
C ::= event CC|CC C C (C) − [C]. The approach
is based on the CRS/ON ERA [16] chronicle description
language that introduces several operators such as:
-
p2
t3
The translation of any chronicle model will consist in
recursively applying the correct pattern to the chronicle model.
For instance, the translation of C2 consists in applying sequence
to C1 first and then divergence. In this way, the LT P N P r
of any chronicle results from the representation of a basic
pattern or the result of a previous combination. The approach
is applied in the context of diagnosability analysis (see V),
therefore the Petri Net represents the chronicle model and the
part of the recognition language that is relevant to diagnosability.
In this work, chronicle model provides the shortest words of
the recognition language that are considered as a faulty (or
normal) behavior of the monitored system.
The modeling of the chronicle recognition is also a challenge
shown in the next section.
sequence: n fully ordered patterns, for example a
sequence of two events
C1: event(a, t1 ) ∧ event(b, t2 ) ∧ t2 − t1 ≥ 4.
divergence: an initial shared pattern precedes n
parallel patterns, for example C2: event(a, t0 ) ∧
event(b, t3 ) ∧ C1 ∧ t1 − t0 ≥ 4 ∧ t3 − t0 ≥ 5.
convergence: n parallel patterns precede a final shared
pattern, for example C3: event(a, t0 )∧event(b, t3 )∧
C1 ∧ t3 − t0 ≥ 2 ∧ t2 − t3 ≥ 5.
disjunction: C1 C2, C1 or exclusive C2
conjunction C1C2, C1 and C2 can occur in any
order
sequence C1C2, C1 then C2
absence (C1) − [C2], C1 without any occurrence of
C2 during the recognition of C1.
For instance, the chronicle (EF ) − [G] corresponds to the
sequence EF (event E followed by event F ) without event G
occurring between E and F . The modeling of the recognition
relies on the modeling of the different operators with Coloured
Petri Nets (CP N ) that extend classical Petri Nets with colors
assigned to the tokens [46]. The places of the net are typed i.e.
a color set is assigned to each place and then a place contains
multisets of tokens. It is then possible that the enabling of
a transition depends on the input marking i.e. on the colors
of the token of the input places. Input and output edges of
transitions are labelled by expressions in which variables of
certain type are involved. The firing of the transition supposes
then a binding of the variables to the colors of the tokens such
that the transition is enabled. The basic principle of chronicle
recognition modeling proposed is that the places are used
to store in a single token a list of all chronicles instances.
A chronicle instance is represented by a list of events, and a
chronicle recognition set by a list of lists of events. The authors
mention that these coloured nets are designed in such a way
that they can be easily composed with others by using place
fusion to consider complex chronicles. Figure 10 presents the
CP N for a simple chronicle corresponding to the occurrence
of events A and B.
18. occurrence of event b, but this will “complete nothing” and
leave unchanged the marking. Now as soon as there is an
event a recognition, it will be completed with each event b
recognition to produce the recognitions of chronicle A B.
B
a
⎥ ⎢
⎢
⎥
⎣ instB ⎦ →⎣
⎦
instB
2 Num
cpt⎤
⎡cpt
[[]]
cpt+1
)⎥
End⎢ ANR(currB , Ea
⎥
→⎢
⎣
instB of two⎦events. By introducing time,
between the occurrence
cpt + 1
the objective is to determine whether durations between events
Success
must be taken into account by the diagnostic tool to improve
When the event is a B occurrence and 2 SuccessA = [],
the overall diagnosability [48]
then the content of place Success (on the right hand side net)
Chronicles can be used to gathered both the by
now evolves to include all new occurrences of AB obtainedknowledge
about the underlying system and aboutA occurrences.
combining the last B event with all previous the faulty executions.
The underlying system is then supposed to behave like a timed
m
modelabM = ⎤ ⎡ ) with E the set of dated event⎤
(E, T
trajectories
⎡
[[]]
of
Start the system (i.e. the system language) and T the set of time
[[]]
⎥
curr dates of the events.
2 SuccessA⎢ currB ⎥ B ⎢
constraints between the occurrence B
⎥
⎥→⎢
⎢
⎣ ANR(instB CPR([[E cpt+1 ]], curr the Supervisory
Success ⎣ instB ⎦ this track we, present the work ofB )) ⎦
To illustrate
b
Num
2 Control cpt
cpt
⎡(DISCO) team of LAAS − CN RS.
⎤
[[]]
In [58], the set of trajectories of the system leading to the
⎥
curr
End⎢
recognition of the chronicleBc is called the recognition language
⎥
→⎢
⎣ ANR(instB , CPR([[E cpt+1 ]], currB )) ⎦
b
L(c). Each chronicle is associated to its observable recognition
cpt + of
language C, that is the set 1 observable projections (i.e. the
Disjunction recognition
Fig.
CPN for recognition
Fig. 10: The CP N 3.model for A Bthe recognition of the AB E. trajectory restricted to the observable events) of any trajectory
of construction of the coloured Petri fault f (i.e. a fault
TheL(c). Each abnormal situation or net for disjunction event
chronicle (directly issued from [18])
Note also that both nets for A and B have variables curr
like in [64] or a fault pattern like in [47] or in nets
is very straightforward: places Start and Success of both [48]) has a
and inst. In order to distinguish them when dealing with
are signatureand both nets work in parallel. We do not go into system
merged, Sig(f ) that is the observable behavior of the
the overall marking, they will be denoted respectively currA ,
further detail as to this operatorchronicle model c(f ) is associated
when the fault occurs. A for brevity reasons.
The and currB , instB .
instA authors define the operational semantics of chronicle
to a fault f when its observable recognition language C is a
recognition modeled using CP N for the different constructions 8 We provide here a simplified definition of complete, while its full f
subset of the fault signature C account the case
7 When A and B are the same name, we deal with “chronicles with
definition includes a mechanism to take into f ⊆ Sig(f ). with repetition,
of the language and prove that the recognition provided by the
repetitions” which are taken into account homogeneously in our processing.
e.g. chronicle AA.
Under the single fault assumption, the authors propose to
CP N is a suitable representation of the recognition for the
check the diagnosability of a fault f relying only on a set of
CRS/ON ERA language. The modeling of the recognition is
chronicles by checking whether two chronicles c(f ) and c(f )
of main interest for formal verification activities in critical and
are exclusive or not. Two chronicles are defined exclusive if
104
large scale systems. The work has been applied in the context
they cannot be recognized with the same flow of events.
of simulation data analysis notably in an airport simulation to
The proposed exclusiveness analysis is performed in the
detect faulty simulation activities [21].
following way:
V. D IAGNOSABILITY A NALYSIS BASED ON CHRONICLES
Check for the non exclusiveness of chronicles c(f1 )
and c(f2 ): if Cf1 ∩ Cf2 = ∅ then f1 and f2 are not
Nowadays, the design of a posteriori tool for the diagnosis
diagnosable.
of a system is no longer imaginable as systems gets more and
Check for the non exclusiveness between a chronicle
more complex. It is imperative to be concerned at the design
c(f ) and a non faulty model of the monitored system
stage of the system diagnosis purposes to achieve. Several
c(f0 ): if Cf ∩ Cf0 = ∅ then f is not diagnosable and
research works tend to characterize and analyze properties the
more precisely f is not detectable.
system should have to make the diagnostic tool efficient on-line
Note that in the case where Cf = Sig(f ) checking for the
to monitor and isolate the faults of the system: among these
exclusiveness allows to conclude on the diagnosability property.
properties, diagnosability is the most studied one. Diagnosability
This approach has been applied in the context of the
is closely related to the capability of the monitoring to record
W S − DIAM ON D european project with the objective to
observations which are necessary to determine the failure causes
develop self-healing Web Services [25]. More recently, [37]
within the system with certainty. In case of dynamic systems,
propose a fully automated and formal method to perform these
diagnosability analysis usually consists in analyzing, off-line,
exclusiveness tests. This method relies on three main steps as
the system trajectories with respect to their observability to
illustrated on Figure 11:
determine whether the corresponding diagnostic tool will be
During the translation step each chronicle model is
able to diagnose the faults online. In the context of discretemodeled into a Labeled Time Petri Net with Priorities
event systems (DES) the work developed by [64] introducing
(LTPNPr). according the method we have depicted in
the diagnoser notion is a reference.
section IV-C.
So, an intuitive way to tackle diagnosability is to answer the
Then a Product step aims to construct from the
question: is there a diagnosis function which from an observable
LT P N P r model of each translated chronicle a
behaviour of the system can tag the observable behaviour with
unique LT P N P r (called product) that models the
a label that is either, ”normal”, ”faulty” or ”ambiguous” and
possible common behaviors with synchronized events.
this in a bounded delay?.
For Timed Discrete Event Systems (TDES). The aim is to
Indeed, the exclusiveness test aims to check that
improve the characterization of diagnosability for a discrete
the chronicles cannot be recognized by a common
event system by taking into account the notion of finite durations
trajectory of events.
19. General principle of exclusiveness analysis based on
chronicles
The system (E , T )
Chronicles Database
Projection
Data
Translation
(Eobs , Tobs )
LTPNPrs
Product
LTPNPr
Steps
Generation
SCG
Analysis
Exclusiveness?
Analysis
Results
LTPNPr: Labeled Time Petri Nets with Priorities
SCG: State Class Graph
Fig. 11: General principle of exclusiveness analysis based on
chronicles
-
Finally, an Exclusiveness analysis step performs
the exclusiveness tests. The exclusiveness analysis
must deal with an important number of trajectories
that means chronicle instances that may induce the
chronicle recognition. These chronicles instances
correspond to the marked behaviors of the Petri Net.
By introducing time intervals between two events,
the state-space associated to the set of possible
trajectories is usually infinite. In order to perform
any diagnosability analysis, it is then necessary
to use an abstraction of this state-space as the
complete enumeration of the possible instances of
each chronicle is not realistic. The authors propose
to consider a time abstraction through the State
Class Graph (SCG) of Time Petri Nets [9] and
to perform the exclusiveness analysis on this time
abstraction. Given SCf = {c1 (f ), . . . , cn (f )} a set
of chronicles associated to a fault f and a set of
chronicles SCf = {c1 (f ), . . . , cn (f )} associated
to a fault f . As previously explained checking the
non exclusiveness between at least one element of
SCf and one element of SCf allows to conclude to
the non diagnosability of the faults f and f . Then,
in a first stage from the SCG the set of trajectories
leading to the recognition of the two chronicles noted
WOK . WOK is the set of trajectories leading to the
marking of the pok places of the chronicle models
(see section IV-C). Then, each of these trajectories is
compared to the set of observable trajectories that the
system can generate (without taking the events date
into account) noted WOBS . If WOK WOBS = ∅
then the two chronicles are exclusive. If furthermore
the faulty behavior associated to f (resp f ) is
totally recognized by SCf (resp SCf ) then the
system is diagnosable. If WOK WOBS = ∅ the two
chronicles are not exclusive and the two faults f and
f are not diagnosable. The solution of the inequalities
system TOBS ∧ Tc gives the precise intervals where
the two chronicles are not exclusive. With Tc the time
constraints on the paths leading to the recognition of
both chronicles and TOBS the restriction of the time
constraints of the chronicle to the observable events.
To validate the proposed approach a diagnosability checker tool
has been developed, based on TINA (TIme Petri Net Analyzer,
(http://www.laas.fr/tina)) [8].
The main difficulty of this approach is the combinatory
explosion intrinsic of the analysis. Therefore in [38] the authors
propose an approach based on Petri nets unfolding to face
this problem by benefiting of a representation as a partial
order of events. Current work address the problem of checking
diagnosability of patterns in discrete-event systems. A pattern
is a partial order of observable/no observable events and is then
8 / 20
very similar to chronicles.
VI. C ONCLUSION
This paper provides an overview of research investigated in
the field of chronicle recognition. As we have tried to point out,
the chronicle recognition is used in a wide range of applications
from medical diagnosis to reconfiguration of communication
protocols. One major advantage of chronicle recognition notably
for diagnosis purposes is the efficiently on line. The approaches
developed to translate chronicles into others formalism like
Petri nets allow to use model checking technics for formal
verification purposes. Moreover, the design of the chronicle
takes benefits of a large amount of works developed in the data
mining community.
This picture allows us to identify a number of research issues.
We can highlight for instance the chronicle design recovery
issued from the diagnosability analysis based on chronicles. The
problem of on line learning of chronicles is also an interesting
challenge as the problem of the chronicle distribution in the
context of distributed applications.
VII. ACKNOWELEDGEMENT
I want to strongly thank the members of the Diagnosis and
Supervisory Control (DISCO) team of the LAAS-CNRS, the
colleagues of the DAISY project and of the Post Graduate
Program with Universidad de los Andes de Merida - Venezuela
involved in many works presented in this paper.
R EFERENCES
[1] R. Agrawal and R. Srikant, “Mining sequential patterns,” in Proceedings
of the Eleventh International Conference on Data Engineering, ser. ICDE
’95, Washington, DC, USA, 1995, pp. 3–14.
[2] J. Aguilar, K. Bousson, C. Dousson, M. Ghallab, A. Guasch, R. Milne,
C. Nicol, J. Quevedo, and L. Trav´ -Massuy` s, “Tiger: real-time situation
e
e
assessment of dynamic systems,” LAAS-CNRS, Toulouse, France,
Technical report, 1994.
[3] J. Aguilar, A. Subias, and L. Trav´ -Massuy` s, “Situation assessment
e
e
in autonomous systems,” in Global Information Infrastructure and
Networking Symposium ( GIIS ), Choroni (Venezuela), december 2012.
[4] J. Aguilar, A. Subias, L. Trave-Massuyes, and K. Zouaoui, “A chronicle
learning approach for self-adapting strategies in collaborative communicating systems,” LAAS-CNRS, France, Technical Rapport 1xs1698,
October 2011.
[5] J. F. Allen, “Maintaining knowledge about temporal intervals,” Communications of the ACM, vol. 26(11), pp. 832 – 843, 1983.
[6] ——, “Towards a general theory of action and time,” Artificial Intelligence,
vol. 23(2), pp. 123–154, 1984.
20. [7] B. Berthomieu, F. Peres, and F. Vernadat, “Bridging the gap between
timed automata and bounded time Petri nets,” in Proceedings of Formal
Modeling and Analysis of Timed Systems (FORMATS’06), LCNS 4202,
2006.
[8] B. Berthomieu, P. Ribet, and F. Vernadat, “The tool tina – construction
of abstract state spaces for Petri nets and time Petri nets,” International
Journal of Production Research, vol. 42, no. 14, 2004.
[9] B. Berthomieu and F. Vernadat, “State class constructions for branching
analyzis of systems,” in Tools and Algorithms for the Construction and
Analysis of Systems (TACAS’03), Warsaw, Poland, 2003.
[10] O. Bertrand, P. Carle, and C. Choppy, “Chronicle modelling using
automata and coloured Petri nets,” in 18th Internation Workshop on
Principles of Diagnosis (DX’07), Nashville, TN, USA, 2007, pp. 243–
248.
[11] ——, “Towards a coloured Petri nets semantics of a chronicle language
for distributed simulation processing,” in In CHINA 2008 Workshop
(Concurrency metHods: Issues aNd Applications), 2008, pp. 105–119.
[12] A. Boufaied, A. Subias, and M. Combacau, “Chronicle modeling by Petri
nets for distributed detection of process failures,” in IEEE International
conference on systems, man and cybernetics (SMC’02), Hammamet,
Tunisie, 2002, pp. 243–248.
[13] ——, “Distributed fault detection with delays consideration,” in 8th World
Multi-Conference on Systemics, Cybernetics and Informatics (SCI’2004),
Orlando,USA, july 2004, pp. 135–140.
[14] ——, “The distributed time constraints verification modelled with time
Petri nets,” in 17th IMACS World Congress on Scientific Computation,
Applied Mathematicsand Simulation (IMACS05), Paris, France, july 2005.
[15] J. Cardoso, R. Valette, and D. Dubois, “Fuzzy Petri nets: an overview,”
in 13th IFAC World congress : IFAC’96, vol. J, San Francisco, USA, july
1996, pp. 443–448.
[16] P. Carle, P. Benhamou, M. Ornato, and F. Dolbeau, “Building dynamic
organizations using intentions recognition,” Office national d’tudes et
de recherches arospatiales, Chˆ tillon, France, Tech. Rep. ONERA-TP a
99-136, 1998.
[17] P. Carle, C. Choppy, R. Kervac, and A. Piel, “Handling breakdowns in
unmanned aircraft systems,” in 18th International Symposium on Formal
Methods - Doctoral Symposium, Paris,France, august 2012, pp. 286–291.
[18] P. Carle, C. Choppy, and R. Kervarc, “Behaviour recognition using
chronicles,” in Proceedings of the 2011 Fifth International Conference on
Theoretical Aspects of Software Engineering, ser. TASE ’11. Washington,
DC, USA: IEEE Computer Society, 2011, pp. 100–107.
[19] P. Carle, C. Choppy, R. Kervarc, and A. Piel, “Safety of Unmanned
Aircraft Systems Facing Multiple Breakdowns,” in 1st French
Singaporean Workshop on Formal Methods and Applications (FSFMA
2013), ser. OpenAccess Series in Informatics (OASIcs), C. Choppy
and J. Sun, Eds., vol. 31. Dagstuhl, Germany: Schloss Dagstuhl–
Leibniz-Zentrum fuer Informatik, 2013, pp. 86–91. [Online]. Available:
http://drops.dagstuhl.de/opus/volltexte/2013/4093
[20] G. Carrault, M.-O. Cordier, R., M. Garreau, J. Bellanger, and A. Bardou,
“A model-based approach for learning to identify cardiac arrhythmias,”
in AIMDM, 1999, pp. 165–174.
[21] C. Choppy, O. Bertrand, and P. Carle, “Coloured Petri nets for chronicle
recognition,” in Proceedings of the 14th Ada-Europe International
Conference on Reliable Software Technologies, ser. Ada-Europe ’09.
Berlin, Heidelberg: Springer-Verlag, 2009, pp. 266–281.
[22] M.-O. Cordier and C. Dousson, “Alarm driven monitoring based on
chronicles,” in 4th Sumposium on Fault Detection Supervision and Safety
for Technical Processes (SafeProcess), Budapest, Hungary, june 2000,
pp. 286–291.
[23] M.-O. Cordier, X. L. Guillou, S. Robin, L. Roz´ , and T. Vidal, “Distributed
e
chronicles for on-line diagnosis of web services,” in 18th International
Workshop on Principles of Diagnosis, G. Biswas, X. Koutsoukos, and
S. Abdelwahed, Eds., May 2007, pp. 37–44.
[24] D. Cram, B. Fuchs, Y. Pri, and A. Mille, “Dcouverte complte et
interactive de motifs temporels avec contraintes numriques partir
de squences d’vnements,” in Atelier ”Fouille de donnes temporelles
et analyse de flux de donnes”, EGC2009, 2009. [Online]. Available:
http://liris.cnrs.fr/publis/?id=3786
[25] W.-D. D4.2, “Specification of diagnosis algorithms for web-services phase 1,” WS-DIAMOND European project, Tech. Rep., September 2006.
[26] C. Diop, E. Exposito, C. Chassot, and K. Drira, “Autonomic service
bus,” in 6th International Conference of Interoperability for Enterprise
Systems and Applications (I-ESA12)Workshop Factories of the Future
(FoF), Valencia - Spain, march 2012.
[27] C. Dousson, “Alarm driven supervision for telecommunication networks:iion line chronicle recognition,” Annales des T´ l´ communications, vol. 51
ee
(9-10), pp. 501–508, 1996.
[28] C. Dousson and T. V. Duong, “Discovering chronicles with numerical
time constraints from alarm logs for monitoring dynamic systems,” in
IJCAI 99: Proceedings of the Sixteenth International Joint Conference on
Artificial Intelligence, San Francisco, CA, USA, June 1999, pp. 620–626.
[29] C. Dousson, P. Gaborit, and M. Ghallab, “Situation recognition: representation and algorithms,” in IJCAI: International Joint Conference on
Artificial Intelligence, Chamb´ ry, France, august 1993, pp. 166–172.
e
[30] E. Exposito, C. Chassot, and M. Diaz, “New generation of transport
protocols for autonomous systems,” in IEEE Global Telecommunications
Conference (Globecom 2010), Miami, Florida (USA), december 2010.
[31] F. Fessant and F. Cl´ rot, “An efficient som-based pre-processing to
e
improve the discovery of frequent patterns in alarm logs,” in DMIN,
2006, pp. 276–282.
e
[32] F. Fessant, F. Cl´ rot, and C. Dousson, “Mining of an alarm log to improve
the discovery of frequent patterns,” in Industrial Conference on Data
Mining, 2004, pp. 144–152.
[33] M. W. Floyd and B. Esfandiari, “A case-based reasoning framework for
developing agents using learning by observation,” in ICTAI, 2011, pp.
531–538.
[34] M. Floyd, M.Bicakci, and B. Esfandiari, “Case-based learning by
observation in robotics using a dynamic case representation,” in FLAIRS
Conference, 2012.
[35] B. Frances, “Apprentissage de chroniques pour les rseaux de communications autonomes,” LAAS-CNRS, France, Master, Toulouse 3, Paul
Sabatier University, September 2012.
[36] J. Gamper and W. Nejdl, “Abstract temporal diagnosis in medical domains,”
in Artificial Intelligence in Medicine, vol. 10(3), 1997, pp. 209–234.
[37] H. Gougam, A. Subias, and Y. Pencole, “Timed diagnosability analysis
based on chronicles,” in 1IFAC International Symposium on Fault
Detection Supervision and Safety of Technical Processes (SAFEPROCESS’2012), Mexico (Mexique), august 2012, pp. 1256–1261.
[38] ——, “Supervision patterns: formal diagnosability checking by Petri net
unfolding,” in 4th IFAC Workshop on Dependable Control of Discrete
Systems (DCDS 2013), York (UK), sep 2013.
[39] B. Guerraz and C. Dousson, “Chronicles construction starting from the
fault model of the system to diagnose,” in International Workshop on
Principles of Diagnosis (DX04), Carcassonne, France, 2004, pp. 51–56.
[40] X. L. Guillou, M.-O. Cordier, S. Robin, and L. Roz´ , “Chronicles for
e
on-line diagnosis of distributed systems,” in 18th European Conference on
Artificial Intelligence, ECAI08, Patras, Greece, july 2008, pp. 194–198.
[41] T. Guyet and R. Quiniou, “Mining temporal patterns with quantitative
intervals,” in ICDM Workshops, 2008, pp. 218–227.
[42] ——, “Extracting temporal patterns from interval-based sequences,” in
IJCAI, 2011, pp. 1306–1311.
[43] I. J. Haimowitz and I. S. Kohane, “Automated trend detection with
alternate temporal hypotheses,” in In Proceedings of the International
Joint Conference on Artificial Intelligence IJCAI, Chamb´ ry, France,
e
1993, pp. 146–151.
[44] F. Heintz, “Chronicle recognition in the WITAS UAV project a
preliminary report,” in Swedish AI Society Workshop (SAIS2001), 2001.
[45] IBM Corp., An architectural blueprint for autonomic computing.
USA: IBM Corp., Oct. 2004. [Online]. Available: www3.ibm.com/autonomic/pdfs/ACBP2 2004-10-04.pdf
[46] K. Jensen, “An introduction to the practical use of coloured petri nets,”
in Petri Nets (2), 1996, pp. 237–292.
[47] T. J´ ron, H. Marchand, S. Pinchinat, and M.-O. Cordier, “Supervision
e
patterns in discrete event systems diagnosis,” in WODES06: Workshop
on Discrete Event Systems, 2006.
[48] A. Khoumsi and Ou´ draogo, “Diagnosis of faults in real-time discrete
e
event systems,” in SAFEPROCESS’09, Jun 2009, pp. 1557–1562.
[49] S. Kocksk¨ mper and B. Neumann, “Extending process monitoring
a
by event recognition,” in In Proceedings of the Second International
Conference on Intelligent Systems Engineering (ISE’94), 1994.
[50] T. Kohonen, “Self-organized formation of topologically correct feature
maps,” Biological Cybernetics, vol. 43, no. 1, pp. 59–69, 1982.
[51] R. Kowalski and M. Sergot, “A logic-based calculus of events,” New
Generation Computing, vol. 4(1), pp. 67–95, 1986.
[52] P. Laborie and J.-P. Krivine, “Automatic generation of chronicles and its
application to alarm processing in power distribution systems,” in 8th
international workshop of diagnosis (DX97), Mont Saint-Michel, France,
1997.
21. [53] H. Mannila, H. Toivonen, and A. I. Verkamo, “Discovery of frequent
episodes in event sequences,” Data Mining and Knowledge Discovery,
vol. 1, pp. 259–289, 1997.
[54] E. Mayer, “Inductive learning of chronicles,” in European Conference
on Artificial Intelligence, 1998, pp. 471–472.
[55] J. McCarthy and P. J. Hayes, “Some philosophical problems from the
standpoint of artificial intelligence,” Machine Intelligence, vol. 4, 1969.
[56] B. Morin and H. Debar, “Correlation on intrusion: an application of
chronicles,” in 6th International Conference on recent Advances in
Intrusion Detection RAID, Pittsburgh, USA, september 2003.
[57] K. N¨ kel, “Temporal matching: Recognizing dynamic situations from
o
discrete measurements,” in In Proceedings of the International Joint
Conference on Artificial Intelligence IJCAI, Detroit, Michigan USA,
August 1989, pp. 1255 –1260.
[58] Y. Pencol and A. Subias, “A chronicle-based diagnosability approach for
discrete timed-event systems: Application to web-services,” Journal of
Universal Computer Science, vol. 15, no. 17, pp. 3246–3272, 2009.
[59] D. F. P.M. Merlin, “Recoverability of communication protocols: Implication of a theoretical study,” IEEE Trans. Comm., vol. 24, no. 9, pp.
1036–1043, 1976.
[60] D. V. Pynadath and M. P. Wellman, “Accounting for context
in plan recognition, with application to traffic monitoring,” in
Proceedings of the Eleventh conference on Uncertainty in artificial
intelligence, ser. UAI’95. San Francisco, CA, USA: Morgan
Kaufmann Publishers Inc., 1995, pp. 472–481. [Online]. Available:
http://dl.acm.org/citation.cfm?id=2074158.2074212
[61] D. Racoceanu, Z. Zerhouni, and N. Addouche, “Modular modeling and
analysis of a distributed production system with distant specialized maintenance,” in IEEE International conference on robotics and automation,
Washington DC, USA, May 2002, pp. 4046–4052.
[62] N. Rota and M. Thonnat, “Activity recognition from video sequences
using declarative models,” in 14th International Workshop on Principles
of Diagnosis (DX00), Morelia, Michoacen, Mexico, June 2000.
[63] R. Saddem, A. Toguyeni, and T. Moncef, “Consistency’s checking of
chronicles’ set using time Petri nets,” in Control Automation (MED),
2010 18th Mediterranean Conference on. IEEE, 2010, pp. 1520–1525.
[64] M. Sampath, R. Sengputa, S. Lafortune, K. Sinnamohideen, and
D. Teneketsis, “Diagnosability of discrete-event systems,” IEEE Transactions on Automatic Control, vol. 40, pp. 1555–1575, 1995.
[65] A. Subias, E. Exposito, C. Chassot, L. Trav´ -Massuy` s, and K. Drira,
e
e
“Self-adapting strategies guided by diagnosis and situation assessment
in collaborative communicating systems,” in International Workshop on
Principles of Diagnosis (DX 10), Portland (USA), 2010.
[66] N. VanWambeke, F. Armando, C. C. C, and E. Exposito, “A modelbased approach for self-adaptive transport protocols,” Elsevier Computer
Communications, Special issue on end-to-end support over heterogeneous
wired and wireless network, vol. 31(11), pp. 2699–2705, July 2008.
[67] N. VanWambeke, E. Exposito, and M. Diaz, “Transport layer qos
protocols: the micro-protocol approach,” in 1st OpenNet Workshop,
Brussels, Belgium, March 2007.
[68] A. Vautier, M.-O. Cordier, and R. Quiniou, “Extension des bases de
donn´ es inductives pour la d´ couverte de chroniques,” in EGC, 2005, pp.
e
e
421–432.
[69] J. Vizcarrondo, J. Aguilar, E. Exposito, and A. Subias, “Armiscom:
Autonomic reflective middleware for management service composition,”
in Global Information Infrastructure and Networking Symposium ( GIIS
), Choroni (Venezuela), december 2012.
[70] J. Vizcarrondo, J. Aguilar, A. Subias, and E. Exposito, “Cronicas
distribuidas para el reconocimiento de fallas en la composicin de servicios
web,” in XXXIX Latin American Computing Conference (CLEI 2013),
Puerto Azul(Venezuela), october 2013.
[71] W. Wang, T. Guyet, R. Quiniou, M.-O. Cordier, and F. Masseglia, “Online
and adaptive anomaly detection: detecting intrusions in unlabelled audit
data streams,” in EGC, 2009, pp. 457–458.
22. Integration of Different Facets of Diagnosis from
Control and AI
L. Trav´ -Massuy` s
e
e
CNRS, LAAS, 7, avenue du Colonel Roche, F-31400 Toulouse, France
Univ of Toulouse, LAAS, F-31400 Toulouse, France
Email: louise@laas.fr
Abstract—Diagnosis is a rich scientific domain driven by the
goal of identifying the root cause of a failure, problem, or disease
from the symptoms arising from selected measurements, checks
or tests. The different facets of the diagnosis problem and the wide
spectrum of classes of systems make this problem interesting to
several communities and call for bridging theories from different
fields. This paper presents diagnosis theories proposed by the
Control and the AI communities and exemplifies how they can
be synergically integrated to provide better diagnostic solutions
and to interactively contribute in fault management architectures.
I.
I NTRODUCTION
The goal of diagnosis is to identify the possible causes
explaining observed symptoms. A set of concomitant tasks
contribute to this goal, in particular :
•
fault detection, which aims at discriminating normal
system states from abnormal ones, i.e. states which
result from the presence of a fault,
•
fault isolation, also called fault localization, whose
goal is to point at the faulty components of the system,
•
fault identification, whose output is the type of fault
and possibly the model of the system under this fault.
In front of the diversity of systems and different views
of the above problems, several scientific communities have
addressed these tasks and contribute with a large spectrum
of methods. The Signal Processing, Control and Artificial
Intelligence (AI) communities are on the front.
Diagnosis works from the signals that permit efficient fault
detection towards the upper levels of supervision that call for
qualitative interpretations. Signal processing provided specific
contributions in the form of statistic algorithms for detecting
changes in signals, hence detecting faults. This track has been
surveyed in several reference books and papers [1], [2], [3],
[4], [5] and remains out of the scope of this paper.
Interfaces between continuous signals and their abstract
interpretations, in symbolic or event-based form, implement
the qualitative interpretations of the signals that are required
for supervision. To do that, discrete formalisms borrowed
from Artificial Intelligence find a natural link with continuous
models from the Control community. These two communities
have their own model-based diagnosis track :
•
the FDI (Fault Detection and Isolation) track, whose
foundations are based on engineering disciplines, such
as control theory and statistical decision making,
•
the DX (Diagnosis) track, whose foundations are
derived from the fields of logic, combinatorial optimization, search algorithms and complexity analysis.
In the last decade, there has been a growing number of
researchers in both communities, who tried to understand and
incorporate approaches from the FDI and DX fields to build
better, more robust and effective diagnostic systems. In this
paper, the concepts and results of the FDI and DX tracks
are put in correspondence and the lessons learned from this
comparative analysis are pointed out.
Data-based diagnosis approaches based on machine learning techniques are present in both the Control and AI communities and complement model-based approaches to provide
solutions to a variety of diagnostic problems where difficulty
arises from the scarce nature of the instrumentation or, conversely, from the massive amounts of data to be interpreted for
the emergence of hidden knowledge. Interesting bridges can be
foreseen between model-based and data-based approaches and
these are illustrated in this paper with the problem of learning
the models that support model-based diagnosis reasoning.
Other bridges can be found when considering that diagnosis
is not a goal per se but a component in fault management architectures. It takes part in the solutions produced for tasks such
as design, failure-mode-and-effects analysis, sensor placement,
on-board recovery, condition monitoring, maintenance, repair
and therapy planning, prognosis. The contribution of diagnosis
in such architectures means close links with decision tasks such
as control and planning and calls for innovative integrations.
In this paper, different facets of diagnosis investigated in
the Control or the AI fields are discussed. While [6], [7], [8]
provide three interesting surveys of the different approaches
that exist in these fields, this paper aims at reporting the
works that integrate approaches of both sides, hence creating
”bridges”.
The paper is organized as follows. After the introduction
section, section II first presents a brief overview of the approaches proposed by the model-based diagnosis communities,
FDI and DX, in subsections II-A and II-B, respectively. Although quite commonplace, this overview is necessary because
it provides the basic concepts and principles that form the foundations of any diagnosis method. It is followed by subsection
II-C that compares the concepts and techniques used by the
FDI and DX communities and presents the lessons learned
from this comparative analysis. Section III is concerned with
the trends that integrate and take advantage of techniques
from both sides, in particular causal model-based diagnosis in
23. subsection III-A and diagnosis of hybrid systems in subsection
III-B. Section IV then raises the problem of obtaining the
models supporting diagnosis reasoning and discusses bridges
that can contribute to learning them in an automated manner.
Section V widens the scope of diagnosis and is concerned with
diagnosis as a component of fault management architectures,
discussing several links with control and planning. Finally,
section VI concludes the paper.
II.
The diagnosis principles are the same, although each community has developed its own concepts and methods, guided
by different modelling paradigms. FDI relies on analytical
models, linear algebra, and non linear system theory whereas
DX takes its bases in logic formalisms. In the 2000s, catalyzed
by the BRIDGE group ”Bridging AI and Control Engineering
model-based diagnosis approaches ” [9] within the Network
of Excellence MONET II [10] and its French counterpart, the
IMALAIA group ”Int´ gration de M´ thodes Alliant Automae
e
tique and IA” supported by GDR MACS [11], GDR I3 [12],
as well as AFIA [13], there were more and more researchers
who tried to understand and synergistically integrate methods
from the two tracks to propose more efficient diagnostic
solutions. This collaboration resulted in the organization of
several events :
•
BM : dx/dt = f (x(t), u(t), θ)
OM : y(t) = g(x(t), u(t), θ).
DX AND FDI MODEL - BASED DIAGNOSIS BRIDGE
The FDI and DX streams both approach the diagnosis
problem from a system point of view, hence resulting in
large overlaps, including the name of the tracks: Model-Based
Diagnosis (MBD).
•
between system inputs and outputs, i.e. the set of measurable
variables Z, as well as the internal states, i.e. the set of
unknown variables X. The variables z ∈ Z et the variables
x ∈ X are functions of time. The typical model may be
formulated in the temporal domain, then known as a statespace model of the form:
a BRIDGE Workshop in 2001 in the framework of
DX’01, 12th International workshop on Principles of
Diagnosis, Sansicario, Via Lattea, Italy, 5-9 Mars 2001
[14],
the co-location of the two main events of the FDI and
the DX communities, namely the Symposium IFAC
Safeprocess 2003 and the International Workshop
”Principles of Diagnosis” DX 2003, in Washington
DC (USA) in June 2003 with a BRIDGE Workshop
in the form of a join day.
This events were followed by the publication of a special
issue of the IEEE SMC Transactions, Part B, on the topic
Diagnosis of Complex Systems: Bridging the methodologies of
the FDI and DX Communities in 2004 by [15].
The Bridge track was launched and is still active today.
Lets’s mention the two invited sessions ”AI methods for Modelbased Diagnosis” and ”Bridge between Control Theory and
AI methods for Model-based Diagnosis”, recently organized in
the framework of the 7th IFAC Symposium on Fault Detection,
Supervision and Safety of Technical Processes Safeprocess’09,
Barcelona, Spain, 30 July-3 August 2009 [16].
The next subsections first summarize the foundations of the
FDI and DX methods, then proceed to a comparative analysis
that allows us to draw some practical assessments in the form
of lessons learned.
A. Brief overview of FDI approaches
The detection and diagnosis methods of the FDI communauty rely on behavioral models that establish the constraints
(1)
where x(t) ∈ nx is the state vector, u(t) ∈ nu is the
input vector and y(t) ∈ np is the output vector. BM is
the behavioral model and OM is the observation model. The
whole model is noted SM (z, x). The equations of SM (z, x)
may be associated to components but this information is not
represented explicitly. The models can also be formulated in
the frequency domain (transfer functions in the linear case).
Models are used in three families of methods:
•
the methods based on parameter estimation that focus
on the value of parameters as representing physical
features of the system,
•
the methods based on state estimation that rely on the
estimation of unknown variables,
•
the methods based on the parity space that rely on the
elimination of unknown variables.
The books [17], [18], [19], [20] provide excellent surveys,
which cite the original papers that the reader is encouraged
to consult. The equivalence between observers, parity and
paramater estimation has been proved in the linear case [21].
The concept central to FDI methods is the concept
of residual and one of the main problems is to generate
residuals. Let’s consider the model SM (z, x) of a system
in the form (1). SM (z, x) is said to be consistent with an
observed trajectory z, or simply consistent with measurements
z, if there exists a trajectory of x such that the equations of
SM (z, x) are satisfied.
Definition 1 (Residual generator for SM (z, x)): A sys˜
tem that takes as input a sub-set of measured variables Z ⊆ Z
and generates as output a scalar r,is a residual generator for
the model SM (z, x) if for all z consistent with SM (z, x),
we have limt→∞ r(t) = 0.
When the system model is consistent with measurements,
the residuals tend to zero as t tends to infinity, otherwise
some residuals may be different from zero. The residuals are
often optimized to be robust to disturbancies [22] and to take
into account uncertainties [23]. The evaluation of residuals
and assigning them a boolean value (0 or non 0) generally
calls for statistical decision techniques [19].
The methods based on parameter estimation are used for
linear as well as non linear systems [24]. Fault detection
is achieved by comparing the estimated parameter values to
their nominal values. With these methods, fault detection,
isolation, and identification are achieved at once, provided that
model parameters can be put in correspondence with physical